|
Article Excerpt
The scheduling of nursing staff is a long-standing problem with myriads of research models published by academia. The exploratory research that we discuss examines the models that academia has produced and the models that hospitals have actually used. We use data from many sources, including research articles, e-mail and telephone surveys, an industry database, and a software source catalog. Only 30 percent of systems that research articles discuss are implemented, and there is very little academic involvement in systems that third-party vendors offer. We examine causes for the research-application gap and discuss directions for future academic research to make it more applicable.
Key words: OR/MS implementation; personnel; manpower planning; hospitals. History: This paper was refereed.
**********
Nurse scheduling or rostering is the assignment of nurses to days and shifts over a specific scheduling period. Objectives of the scheduling task include minimizing staff to avoid wasted effort, while also ensuring sufficient staffing to provide adequate patient care and ensure service continuity. Any proposed scheduling schema must also satisfy organizational, legislative, and union policies (Sitompul and Randhawa 1990). In addition to these goals and constraints, effective nurse scheduling is critical to staff morale and directly impacts both patient care and nurse retention (Silvestro and Silvestro 2000).
Nurse scheduling is part of the larger capacity-planning problem, which involves staffing (how many nurses are needed, when, and where) and scheduling (which determines when and where each nurse works). These are part of an interrelated, hierarchical problem. The staffing problem is generally solved first and involves forecast of demands, acuity of care forecasting, and integration with nursing availability and skills. Once a staffing plan is finalized, a schedule is constructed that informs individual nurses and nurse managers of who is assigned when and where. This research considers only the scheduling aspect of the larger problem.
The nursing shortage further complicates the nurse-scheduling problem. According to estimates by the Joint Commission on Accreditation of Healthcare Organizations (2005), there were 126,000 unfilled nursing positions in the United States in 2005; in addition, there will be 400,000 fewer nurses than are needed in 2020. The nurse-shortage problem is a global problem (World Health Organization 2000, Pan American Health Organization 2002). This global shortage of nursing staff adds another constraint to the development of nurse-scheduling systems. In addition to the optimization goals and constraints we described in the first paragraph, an efficient and efficacious nurse-scheduling system must satisfy the following conditions: (1) schedules should not require excess hours, (2) scheduling time required by nurse managers must be minimized, and (3) schedules must be acceptable to nurses such that they enhance retention of this already scarce resource.
Management science, operations research, and computer science have offered many solutions to the employee-scheduling problem. For over 40 years, academic literature has addressed nurse scheduling specifically. There are several excellent review articles that summarize this literature (Choi et al. 1991, Siferd and Benton 1992, Hung 1995, Cheang et al. 2003, Burke et al. 2004b). The research articles provide a wide variety of solutions, incorporating almost every conceivable work environment and constraint, including the nuances of adjusting a roster when a scheduled nurse fails to appear. Solutions range from simple algorithms to complex artificial intelligence and decision-support systems. What is unclear is if and how these solutions are used in practice.
There is great potential for improving the use of these solutions. In 2003, the International Council of Nurses (2002-2003) estimated that there were 12million nurses worldwide, and in 2005, the American College of Healthcare Executives (2005) estimated that 2.3 million were in the United States. Simulating a pencil-and-paper scheduling solution (Hung 1991), we estimate that even a 10 percent time-savings improvement in the scheduling task in the United States could save 130,000 nurse-manager hours per year. This represents about 90 full-time-equivalent (FTE) nurse managers. This estimate includes only time savings and does not consider other benefits that would accrue from using a management science approach to the scheduling problem. Admittedly, there are many commercial vendors of computerized scheduling techniques that could or do reduce scheduling time. That market is huge. We estimate projected costs to hospitals in the United States for nurse-scheduling software in 2005 at $156 million. This does not include associated training and maintenance costs. Part of our research will explore the extent to which these software firms utilize academic solutions.
One might expect that there would be some gap in the application of academic solutions to practice. This gap has been evident in production-scheduling techniques (King 1976) and requirements-modeling methods for systems development (Maiden et al. 2005). Our exploratory research investigates how academic research of the nurse-scheduling and rostering problem is used in practice.
In the first section of this paper, we discuss how academic research is used in practice. We then discuss the implementation of nurse-scheduling solutions in the United States and follow with a discussion of the research-application gap in nurse scheduling and an exploration of future research directions.
Current Academic Nurse-Scheduling Research
As we mentioned above, academic research into the nurse-scheduling and rostering problem has a long and well-summarized history. It is not our intent to duplicate that effort. Instead, our research investigates the extent to which academic research that is specifically related to nurse scheduling is transferred to practice.
For the purposes of this research, we defined academic research as research that has been published in an academic journal. While not all hospitals use computers for their scheduling task, almost all of the academic solutions do. We limited the search for published academic research to the period from 1985 to 2005. We chose 1985 as the initial date because inexpensive yet sufficiently powerful PCs were available. It would be unreasonable to expect that a computerized solution would be feasible in practice unless the technology was readily and inexpensively available.
To locate the academic articles, we used standard, academic-library search engines. We used the key words, nurse scheduling and nurse rostering, in the Academic Search Premier, OneTrack (Expanded Academic), Medline, and IEEE Explore search engines and limited publication dates to 1985 through 2005 to represent 20 years of academic research. We did not include Ph.D. dissertations or published papers from symposia. We also did not include anecdotal, referential, tutorial, or case studies. It was important that the research we included not only present a model but also perform a rigorous analysis of the optimality or utility of the technique. In general, we included proceedings articles and used the bibliographies of the aforementioned review articles to validate the library database searches.
To analyze the data, we required access to the full articles or proceedings papers. If we could not obtain such access after performing an extensive online search and enlisting the help of a reference librarian with interlibrary-loan capabilities, we excluded that work. We excluded several white papers because they focused on demonstrating the abilities of an optimization language, not on solving the nurse-scheduling problem. We also excluded articles that modeled the staffing decision (i.e., the number of nurses required).
There are many excellent academic articles on general labor-scheduling techniques that are applicable to a nursing environment. However, we excluded these because our intent was to look at implementation issues with work that was specifically directed toward nurse scheduling. This is not to imply that general scheduling models are not applicable to a nursing environment. However, the effort to move from a general solution to a specific environment is more difficult than moving from a model created for a specific environment to that same environment. In looking ahead to the data in which we were interested, we decided that the extra implementation hurdles faced by general scheduling models (e.g., nursing shortage, regulatory compliance, and improvement in patient-care quality) would complicate the analysis.
Seventy-two research articles met our criteria. Some of these described different aspects of the same scheduling technique or enhancements to a previously published technique; some discussed the same technique for different audiences. The unit of analysis was not an individual research article, but a nurse-scheduling model or application. Combining articles where we deemed it appropriate resulted in 50 nurse-scheduling models. Initially, we gathered data about the techniques from the research articles.
We first examined the articles considering descriptive information such as geographic location, type of platform, and problem type (Table 1).
A personal computer was used in 34 cases (68 percent). In two cases, these were Apple/Mac machines; in two cases, workstations; and in two cases, larger minicomputer or mainframe computers. One case presents a pencil-and-paper system. Nine cases give no indication of the type of computer used.
We categorized...
|
